19 results about "State predictor" patented technology

The invention provides a glyceryl triacetate production parameter control method and system based on reinforcement learning, and relates to the technical field of optimization control methods, and the method comprises the steps: obtaining the production parameters of glyceryl triacetate; based on the kinetic model, constructing a kinetic equation of the esterification reaction of glycerol and acetic acid; constructing a digital twinborn simulation environment of the glyceryl triacetate production process; defining a reinforcement learning framework by taking a digital twin simulation environment as a scene; according to the reinforcement learning framework, taking the Elman neural network as a state predictor, and predicting the state of the glyceryl triacetate; solving the reinforcement learning framework through an MO-MPO algorithm, and determining an optimal control strategy and an optimal production parameter sequence corresponding to the optimal control strategy; judging whether the process capability index is smaller than a process capability index threshold value or not; if yes, continuing to solve the reinforcement learning framework; otherwise, outputting an optimal production parameter sequence; and controlling the optimal production parameter sequence through a PID (Proportion Integration Differentiation) control algorithm.

The invention provides a dynamic risk perception prediction stability control system and method for a cold-chain logistics transport vehicle, and belongs to the technical field of vehicle control. Comprising three core modules: a physical-data mixed residual state predictor, a quantitative stability evaluation module based on a maximum Lyapunov index, and a self-adaptive nonlinear model prediction controller fusing predetermined performance control. And finally, the control instruction is distributed to the four in-wheel motors through a multi-target torque distributor. According to the method, the prediction precision is remarkably improved, continuous risk quantification is realized, and meanwhile, the control performance and the system robustness are synchronously improved.

The invention discloses a communication delay-oriented multi-spacecraft cooperative rendezvous control method, which comprises the following steps of: firstly, respectively designing two delay state predictors for each follower spacecraft, and effectively solving the problem of transmission delay of output states between adjacent spacecrafts induced by communication bandwidth constraints; based on two time delay state predictors, designing a distributed composite state observer to perform real-time estimation on state information, control input, relative state of the observer and the leader and unknown sum disturbance of the leader; and a distributed cooperative rendezvous controller is designed by using an estimated output value of the distributed composite state observer, so that no-error tracking of the follower spacecraft on the state of the leader is realized while the total disturbance is inhibited, and cooperative rendezvous of the whole spacecraft cluster is realized. According to a cooperative control strategy proposed for a spacecraft cluster system, autonomous cooperative rendezvous of spacecraft clusters is realized by using less transmission information with time delay, and the method can be widely applied to other multi-agent formation cooperative systems.

The invention discloses a shield tunnel segment intelligent cooperation method and system based on dynamic damage prediction and self-repairing. The method comprises the steps that digital twin bodies synchronously mapped with a physical tunnel entity are constructed; acquiring strain and temperature data streams, performing data cleaning and feature extraction, and inputting the data streams to the digital twinborn body; real-time simulation is carried out to diagnose the damage state of the physical segment, and a first-level decision node is triggered: a critical state predictor is called, the remaining time T from damage to the critical state is dynamically calculated, a second-level decision node is triggered, and a repair task identifier is generated; the repair strategy optimizer decides an optimal repair scheme and generates a control instruction set, and the mobile repair robot is driven to execute automatic repair operation; and evaluating the repairing efficiency and carrying out self-learning updating. According to the invention, by fusing the real-time sensing data and the multi-factor damage model, accurate diagnosis and dynamic risk prediction of the structural damage are realized, the early warning capability is greatly improved, and dynamic decision and continuous optimization are realized.

The invention discloses an unmanned aerial vehicle reinforcement learning control method and device based on spatial-temporal feature separation, and relates to the technical field of unmanned aerial vehicle control, and the method comprises the steps: separating collected unmanned aerial vehicle sensor data into kinematic features and dynamic features; inputting the kinematic features and the dynamic features into a delay compensation state predictor, and outputting a delay compensation state; performing fluid mechanics constraint verification on the time delay compensation state, and if verification is not passed, performing gradient guide correction on the time delay compensation state; performing feature splicing on the final time delay compensation state and the target task state of the unmanned aerial vehicle, and inputting the spliced features into a strategy network to obtain a control instruction of the unmanned aerial vehicle; the strategy network is obtained through self-adaptive distillation framework training, and the teacher network and the student network are trained in an ideal environment and a delayed environment respectively; the teacher network and the student network are trained by adopting a reinforcement learning algorithm. According to the invention, the attitude recovery time and energy loss of the unmanned aerial vehicle in an interference environment can be reduced.

The present application belongs to the technical field of hot press tank temperature control in composite material forming process, and specifically discloses a kind of inflection point smooth transition method and system of hot press tank temperature control curve, comprising: superimposed test disturbance signal and collected response data before reaching temperature control inflection point;Local time constant and steady-state process gain are identified by constructing mathematical model;The parameter is converted into hard constraint boundary, and the dynamic time sequence transition trajectory sequence is generated by combining the slope before and after inflection point solution;Future desired temperature prediction value is generated using state predictor;Predictive control bias is calculated based on trajectory sequence and prediction value, and trajectory change rate is extracted;Accordingly, proportional integral adjustment component and feedforward compensation component are calculated, and are combined into execution control quantity to drive hot press tank.The present application eliminates temperature overshoot and response delay, and realizes inflection point smooth transition.

The application discloses a kind of intelligent wireless link state prediction methods, first, establish MIMO-OFDM system model;Then propose a kind of Transform time sequence association encoder and the asynchronous advantage actor-critic algorithm (Asynchronous Advantage Actor-Critic, A3C) reinforcement learning framework of adapted online updating;Finally, the corresponding wireless link state predictor is designed.According to the equivalent SINR in system model and the modulation and coding scheme (Modulation and Coding Scheme, MCS) calculation mode simulation verification, via public real MIMO-OFDM operation data set verification, the application has good MCS prediction accuracy, applicable to large-scale MIMO system link prediction.

The application discloses a lithium battery health state and residual service life joint prediction method based on adversarial learning, and belongs to the lithium battery life prediction field. The application firstly performs simple preprocessing on original aging data, then utilizes a shared feature extractor and a specific feature extractor composed of a convolution network module and a residual network module to extract task-shared features and task-specific features of the health state and the residual service life based on adversarial learning, and improves feature discrimination; after the task-shared features and the task-specific features of the health state and the residual service life are fused, the fused features are input into a health state predictor and a residual service life predictor, model negative optimization caused by feature confusion is avoided, and finally, health state and residual service life prediction results are obtained. The application carries out the lithium battery health state and residual service life joint prediction based on adversarial learning, the joint prediction model has low requirements for input, the prediction error is low, and the aging state and the life state of the lithium battery in an actual use scenario can be effectively monitored, and the safety of equipment is improved.

The application discloses a floating wind turbine power control method and device based on wave interval estimation and a medium, relates to the technical field of wind turbine control, and comprises the following steps: acquiring an observation value of a state variable at a current moment and a predicted value of a control input at a previous moment, and determining an estimated value of wave disturbance at the current moment by using an interference interval observer; determining a predicted value of wave disturbance at a next moment based on the estimated values of wave disturbance at the current moment and the previous moment by using a wave disturbance predictor; determining a predicted value of the state variable at the next moment based on the observation value of the state variable at the current moment and the predicted value of wave disturbance at the next moment by using a state predictor, and determining a predicted value of the control input at the current moment by using a time delay controller; and controlling the floating wind turbine at the current moment by using the predicted value of the control input at the current moment. The application realizes accurate compensation and control of wave disturbance.

The invention relates to the technical field of intelligent compensation control, in particular to a linear motor driving system control method based on robust baseline control and adaptive compensation. The method comprises the following steps: establishing a linear motor state space model; a dynamic output feedback baseline controller is obtained by iteratively solving a Riccati equation set based on a disturbance suppression level index, a closed-loop matrix is extracted, and a baseline control signal is output; calculating a steady-state gain and filtering the reference input to generate a steady-state feed-forward control signal; estimating the total uncertainty on line through a state predictor, constructing a projection operator to decompose the total uncertainty into a matched component and an unmatched component, performing output equivalent mapping on the unmatched component, and performing band-limiting processing through a low-pass filter to synthesize a self-adaptive compensation signal; and after the three paths of signals are superposed, the linear motor is driven through amplitude saturation and change rate limitation. According to the method, the robustness and the online compensation capability are fused, and the tracking precision and the anti-interference robustness of the linear motor are remarkably improved.

Systems and methods for determining the total chargeable / dischargeable energy of subsystems of a battery energy storage system (BESS) are disclosed. An iterative process is performed within a dynamic period of time divided into multiple iterations using a neural network model including an energy prediction sub-model and a state prediction sub-model. The energy prediction sub-model outputs chargeable / dischargeable energy for the subsystem of the current iteration. The state prediction sub-model outputs a voltage of the subsystem for the next iteration, a charge rate of the subsystem for the next iteration, a maximum temperature of the subsystem for the next iteration, and a charge rate difference for the next iteration. The total chargeable / dischargeable energy of the BESS subsystem is determined by summing the dischargeable energy determined for each iteration.

The invention relates to the technical field of intelligent compensation control, in particular to a linear time-invariant ball screw driving system control method based on self-adaptive control. The method comprises the following steps: establishing a state space model of a system; designing a baseline controller, obtaining a state feedback gain by solving a linear matrix inequality, and generating a baseline control signal; calculating a steady-state gain, and filtering the reference input signal to generate a steady-state feed-forward control signal; constructing a state predictor, estimating the total uncertainty on line, decomposing the total uncertainty into matching and mismatching signals, generating matching and mismatching compensation signals after gain mapping and low-pass filter band limiting processing, and synthesizing the matching and mismatching compensation signals into self-adaptive compensation signals; and superposing the three paths of signals to obtain a master control signal, and after amplitude saturation and change rate limitation, outputting a system to input a signal driving system. According to the control method provided by the invention, the tracking precision of the ball screw system and the resistance to uncertain factors and external disturbance are effectively improved.

A tire replacement estimation system includes a tire supporting a vehicle. A sensor unit is mounted on the tire and includes a footprint centerline length measurement sensor and a pressure sensor. A processor is in electronic communication with the sensor unit and receives the measured centerline length and the measured pressure. An electronic vehicle network transmits selected vehicle parameters to the processor. A wear state predictor is stored on the processor and receives the measured centerline length, the measured pressure, and the selected vehicle parameters to generate an estimated wear state of the tire. An estimation model is stored on the processor and receives a plurality of estimated wear states of the tire and predicts a future wear state of the tire. The estimation model generates an estimated tire replacement date when the predicted future wear state of the tire exceeds a predetermined wear threshold.

An electromagnetic field generation system and methods are provided. The system includes a plurality of parallel state predictors; and a plurality of switching components; outputs of the plurality of parallel state predictors being provided as inputs to the plurality of switching components. The method for generating an arbitrary electric current includes configuring a cost accumulator of a predictor unit to perform a cost function to obtain predicted accumulated difference of first derivative of an input variable against a reference; configuring the cost accumulator in another predictor unit to perform the cost function as a step function based on the input variable; and configuring the predictor unit to have a variable priority output as a linear function based on the input variable.

In order to solve the problem of product quality detection in an automatic packaging production line, the invention provides a packaging production line detection control method based on an event trigger mechanism, and the method comprises the steps: building a mathematical model of a detection control system through a linear time delay system model with uncertainty; a state predictor based on a system state observer and an event controller are introduced, and then a state feedback controller using the structure is provided. Firstly, a state observer is designed based on the measurement output of the system to estimate the real-time state of the system. Secondly, in consideration of network control signal delay, a state predictor is provided for calculating the future state of the system after time delay. And finally, an event triggering strategy is designed, a state feedback controller is designed, the influence of network signal delay is eliminated, the feasibility of the designed control system is ensured, and the event triggering interval of the system has a right lower bound, that is, Zeno Behavior does not exist. Compared with a control system adopting a time period triggering strategy or a general event triggering strategy, the method has the advantages that the triggering times of the control signals are obviously reduced, the communication resources of a control channel are saved, the action of the controller is triggered as required, the mechanical wear and the energy consumption are reduced, the deviation is corrected in real time, and the control precision is ensured.

A system and method for training a neural network eye state predictor is disclosed. In one example, the method includes feeding a first eye-related observation as input to the eye state predictor to determine a predicted 3D eye state of at least one eye of the subject for a time. The predicted 3D eye state is fed as input to a differentiable predictor to determine a prediction for the at least one eye of the subject for the time. Based on the prediction and at least one of the first eye-related observation and a second eye-related observation, a training loss is determined. The second eye-related observation refers to the at least one eye of the subject for the time. The training loss is used to train the eye state predictor.

This invention belongs to the technical field of autoclave temperature control in composite material molding processes. Specifically, it discloses a method and system for smoothing the inflection point of an autoclave temperature control curve. The method includes: superimposing a test disturbance signal and acquiring response data before reaching the temperature control inflection point; constructing a mathematical model to identify the local time constant and steady-state process gain; converting this parameter into a hard constraint boundary and generating a dynamic time-series transition trajectory sequence by combining the slope calculations before and after the inflection point; generating a predicted future temperature value using a state predictor; calculating the predicted control deviation based on the trajectory sequence and the predicted value, and extracting the trajectory change rate; and calculating the proportional-integral adjustment component and the feedforward compensation component respectively, synthesizing them into an execution control quantity to drive the autoclave. This invention eliminates temperature overshoot and response lag, achieving a smooth inflection point transition.

The application discloses a kind of double-motor servo system's pre-determined performance control method considering external disturbance, the method comprises: the dynamics model of double-motor servo system considering external disturbance is established;Define the total disturbance and state variable of double-motor servo system;According to the total disturbance and state variable of double-motor servo system, the dynamics model of double-motor servo system considering external disturbance is transformed, and the state equation of double-motor servo system is obtained;Radial basis neural network is constructed to estimate load end total disturbance and motor end total disturbance respectively, and the approximation value of load end total disturbance and motor end total disturbance is obtained;According to the prescribed performance function, the prediction tracking error constraint condition is established, and the transformation error is solved according to the smooth strictly increasing function of introduced transformation error;Combined with tracking controller and synchronization controller, the pre-determined performance controller based on state predictor is designed.The application realizes group on the tracking of motor under the premise of guaranteeing double-motor synchronization, and the stability of system is proved by Lyapunov criterion.

The invention discloses an unmanned aerial vehicle control method and system for extreme high-speed flight. A kinetic model with a bottom layer loop, a multi-rate integral state predictor, a variable step size trajectory planning module and a trajectory tracking module are introduced. The bottom loop can enable the input of the dynamic model to be consistent with the control input of the flight firmware, so that the modeling precision is improved, and the difference from reality is reduced. The multi-rate integral state predictor can improve integral precision and avoid prediction saturation distortion. The trajectory planning module calculates a steepest flight trajectory conforming to the physical limit of the unmanned aerial vehicle through a kinetic model and a state predictor, and ensures that a reference trajectory falls within a feasible solution range. The trajectory tracking module solves the state evolution of the unmanned aerial vehicle in a rolling manner according to a dynamic model and a state predictor, and calculates a path error and end point error comprehensive optimal action sequence. The rolling solution and comprehensive optimal design can flexibly adjust actions under limiting conditions and better adapt to a dynamic environment.